open-consul/agent/consul/server_serf.go

408 lines
12 KiB
Go
Raw Normal View History

2013-12-07 01:18:09 +00:00
package consul
2013-12-07 00:54:33 +00:00
2013-12-09 23:29:44 +00:00
import (
"fmt"
"net"
"path/filepath"
"strings"
"time"
"github.com/hashicorp/consul/agent/metadata"
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib"
"github.com/hashicorp/consul/logging"
"github.com/hashicorp/go-hclog"
"github.com/hashicorp/raft"
"github.com/hashicorp/serf/serf"
2013-12-09 23:29:44 +00:00
)
2014-03-20 19:51:49 +00:00
const (
// StatusReap is used to update the status of a node if we
// are handling a EventMemberReap
StatusReap = serf.MemberStatus(-1)
// userEventPrefix is pre-pended to a user event to distinguish it
userEventPrefix = "consul:event:"
// maxPeerRetries limits how many invalidate attempts are made
maxPeerRetries = 6
// peerRetryBase is a baseline retry time
peerRetryBase = 1 * time.Second
2014-03-20 19:51:49 +00:00
)
// setupSerf is used to setup and initialize a Serf
func (s *Server) setupSerf(conf *serf.Config, ch chan serf.Event, path string, wan bool, wanPort int,
segment string, listener net.Listener) (*serf.Serf, error) {
conf.Init()
if wan {
conf.NodeName = fmt.Sprintf("%s.%s", s.config.NodeName, s.config.Datacenter)
} else {
conf.NodeName = s.config.NodeName
if wanPort > 0 {
conf.Tags["wan_join_port"] = fmt.Sprintf("%d", wanPort)
}
}
conf.Tags["role"] = "consul"
conf.Tags["dc"] = s.config.Datacenter
conf.Tags["segment"] = segment
if segment == "" {
for _, s := range s.config.Segments {
conf.Tags["sl_"+s.Name] = net.JoinHostPort(s.Advertise, fmt.Sprintf("%d", s.Port))
}
}
conf.Tags["id"] = string(s.config.NodeID)
conf.Tags["vsn"] = fmt.Sprintf("%d", s.config.ProtocolVersion)
conf.Tags["vsn_min"] = fmt.Sprintf("%d", ProtocolVersionMin)
conf.Tags["vsn_max"] = fmt.Sprintf("%d", ProtocolVersionMax)
conf.Tags["raft_vsn"] = fmt.Sprintf("%d", s.config.RaftConfig.ProtocolVersion)
conf.Tags["build"] = s.config.Build
addr := listener.Addr().(*net.TCPAddr)
conf.Tags["port"] = fmt.Sprintf("%d", addr.Port)
if s.config.Bootstrap {
conf.Tags["bootstrap"] = "1"
}
if s.config.BootstrapExpect != 0 {
conf.Tags["expect"] = fmt.Sprintf("%d", s.config.BootstrapExpect)
}
if s.config.NonVoter {
conf.Tags["nonvoter"] = "1"
}
if s.config.UseTLS {
conf.Tags["use_tls"] = "1"
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
if s.acls.ACLsEnabled() {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
// we start in legacy mode and allow upgrading later
conf.Tags["acls"] = string(structs.ACLModeLegacy)
} else {
conf.Tags["acls"] = string(structs.ACLModeDisabled)
}
var subLoggerName string
if wan {
subLoggerName = logging.WAN
} else {
subLoggerName = logging.LAN
}
// Wrap hclog in a standard logger wrapper for serf and memberlist
// We use the Intercept variant here to ensure that serf and memberlist logs
// can be streamed via the monitor endpoint
serfLogger := s.logger.
NamedIntercept(logging.Serf).
NamedIntercept(subLoggerName).
StandardLoggerIntercept(&hclog.StandardLoggerOptions{InferLevels: true})
memberlistLogger := s.logger.
NamedIntercept(logging.Memberlist).
NamedIntercept(subLoggerName).
StandardLoggerIntercept(&hclog.StandardLoggerOptions{InferLevels: true})
conf.MemberlistConfig.Logger = memberlistLogger
conf.Logger = serfLogger
conf.EventCh = ch
conf.ProtocolVersion = protocolVersionMap[s.config.ProtocolVersion]
conf.RejoinAfterLeave = s.config.RejoinAfterLeave
if wan {
conf.Merge = &wanMergeDelegate{}
} else {
conf.Merge = &lanMergeDelegate{
dc: s.config.Datacenter,
nodeID: s.config.NodeID,
nodeName: s.config.NodeName,
segment: segment,
}
}
// Until Consul supports this fully, we disable automatic resolution.
// When enabled, the Serf gossip may just turn off if we are the minority
// node which is rather unexpected.
conf.EnableNameConflictResolution = false
if !s.config.DevMode {
conf.SnapshotPath = filepath.Join(s.config.DataDir, path)
}
if err := lib.EnsurePath(conf.SnapshotPath, false); err != nil {
return nil, err
}
s.addEnterpriseSerfTags(conf.Tags)
return serf.Create(conf)
}
2014-08-27 01:50:03 +00:00
// userEventName computes the name of a user event
func userEventName(name string) string {
return userEventPrefix + name
}
// isUserEvent checks if a serf event is a user event
func isUserEvent(name string) bool {
return strings.HasPrefix(name, userEventPrefix)
}
// rawUserEventName is used to get the raw user event name
func rawUserEventName(name string) string {
return strings.TrimPrefix(name, userEventPrefix)
2014-08-27 01:50:03 +00:00
}
2013-12-07 00:54:33 +00:00
// lanEventHandler is used to handle events from the lan Serf cluster
func (s *Server) lanEventHandler() {
for {
select {
case e := <-s.eventChLAN:
2013-12-09 23:29:44 +00:00
switch e.EventType() {
case serf.EventMemberJoin:
s.lanNodeJoin(e.(serf.MemberEvent))
s.localMemberEvent(e.(serf.MemberEvent))
case serf.EventMemberLeave, serf.EventMemberFailed, serf.EventMemberReap:
s.lanNodeFailed(e.(serf.MemberEvent))
s.localMemberEvent(e.(serf.MemberEvent))
2013-12-09 23:29:44 +00:00
case serf.EventUser:
s.localEvent(e.(serf.UserEvent))
2017-03-21 23:36:44 +00:00
case serf.EventMemberUpdate:
s.localMemberEvent(e.(serf.MemberEvent))
2014-03-12 19:46:14 +00:00
case serf.EventQuery: // Ignore
2013-12-09 23:29:44 +00:00
default:
s.logger.Warn("Unhandled LAN Serf Event", "event", e)
2013-12-09 23:29:44 +00:00
}
2013-12-07 00:54:33 +00:00
case <-s.shutdownCh:
return
}
}
}
2014-01-09 23:49:09 +00:00
// localMemberEvent is used to reconcile Serf events with the strongly
// consistent store if we are the current leader
func (s *Server) localMemberEvent(me serf.MemberEvent) {
// Do nothing if we are not the leader
if !s.IsLeader() {
return
}
2014-03-20 19:51:49 +00:00
// Check if this is a reap event
isReap := me.EventType() == serf.EventMemberReap
// Queue the members for reconciliation
2014-01-09 23:49:09 +00:00
for _, m := range me.Members {
2014-03-20 19:51:49 +00:00
// Change the status if this is a reap event
if isReap {
m.Status = StatusReap
}
select {
case s.reconcileCh <- m:
default:
}
}
2013-12-09 23:29:44 +00:00
}
// localEvent is called when we receive an event on the local Serf
func (s *Server) localEvent(event serf.UserEvent) {
// Handle only consul events
if !strings.HasPrefix(event.Name, "consul:") {
return
}
switch name := event.Name; {
case name == newLeaderEvent:
s.logger.Info("New leader elected", "payload", string(event.Payload))
// Trigger the callback
if s.config.ServerUp != nil {
s.config.ServerUp()
}
case isUserEvent(name):
event.Name = rawUserEventName(name)
s.logger.Debug("User event", "event", event.Name)
// Trigger the callback
if s.config.UserEventHandler != nil {
s.config.UserEventHandler(event)
}
default:
if !s.handleEnterpriseUserEvents(event) {
s.logger.Warn("Unhandled local event", "event", event)
}
}
}
// lanNodeJoin is used to handle join events on the LAN pool.
func (s *Server) lanNodeJoin(me serf.MemberEvent) {
2013-12-12 00:24:34 +00:00
for _, m := range me.Members {
ok, serverMeta := metadata.IsConsulServer(m)
if !ok || serverMeta.Segment != "" {
2013-12-12 00:24:34 +00:00
continue
}
s.logger.Info("Adding LAN server", "server", serverMeta.String())
// Update server lookup
s.serverLookup.AddServer(serverMeta)
// If we're still expecting to bootstrap, may need to handle this.
if s.config.BootstrapExpect != 0 {
s.maybeBootstrap()
}
2017-03-15 19:26:54 +00:00
// Kick the join flooders.
s.FloodNotify()
}
}
func (s *Server) lanNodeUpdate(me serf.MemberEvent) {
for _, m := range me.Members {
ok, serverMeta := metadata.IsConsulServer(m)
if !ok || serverMeta.Segment != "" {
continue
}
s.logger.Info("Updating LAN server", "server", serverMeta.String())
// Update server lookup
s.serverLookup.AddServer(serverMeta)
}
}
// maybeBootstrap is used to handle bootstrapping when a new consul server joins.
2014-06-18 23:15:28 +00:00
func (s *Server) maybeBootstrap() {
// Bootstrap can only be done if there are no committed logs, remove our
// expectations of bootstrapping. This is slightly cheaper than the full
// check that BootstrapCluster will do, so this is a good pre-filter.
2014-06-18 23:15:28 +00:00
index, err := s.raftStore.LastIndex()
if err != nil {
s.logger.Error("Failed to read last raft index", "error", err)
2014-06-18 23:15:28 +00:00
return
}
if index != 0 {
s.logger.Info("Raft data found, disabling bootstrap mode")
s.config.BootstrapExpect = 0
2014-06-18 23:15:28 +00:00
return
}
// Scan for all the known servers.
2014-06-18 23:15:28 +00:00
members := s.serfLAN.Members()
var servers []metadata.Server
2018-09-20 00:41:36 +00:00
voters := 0
2014-06-18 23:15:28 +00:00
for _, member := range members {
valid, p := metadata.IsConsulServer(member)
2014-06-18 23:15:28 +00:00
if !valid {
continue
}
if p.Datacenter != s.config.Datacenter {
s.logger.Warn("Member has a conflicting datacenter, ignoring", "member", member)
2014-06-18 23:15:28 +00:00
continue
}
if p.Expect != 0 && p.Expect != s.config.BootstrapExpect {
s.logger.Error("Member has a conflicting expect value. All nodes should expect the same number.", "member", member)
2014-06-18 23:15:28 +00:00
return
}
if p.Bootstrap {
s.logger.Error("Member has bootstrap mode. Expect disabled.", "member", member)
2014-06-18 23:15:28 +00:00
return
}
2018-09-20 00:41:36 +00:00
if !p.NonVoter {
voters++
}
servers = append(servers, *p)
2014-06-18 23:15:28 +00:00
}
// Skip if we haven't met the minimum expect count.
2018-09-20 00:41:36 +00:00
if voters < s.config.BootstrapExpect {
2014-06-18 23:15:28 +00:00
return
}
// Query each of the servers and make sure they report no Raft peers.
for _, server := range servers {
var peers []string
// Retry with exponential backoff to get peer status from this server
for attempt := uint(0); attempt < maxPeerRetries; attempt++ {
if err := s.connPool.RPC(s.config.Datacenter, server.Addr, server.Version,
"Status.Peers", server.UseTLS, &structs.DCSpecificRequest{Datacenter: s.config.Datacenter}, &peers); err != nil {
nextRetry := time.Duration((1 << attempt) * peerRetryBase)
s.logger.Error("Failed to confirm peer status for server (will retry).",
"server", server.Name,
"retry_interval", nextRetry.String(),
"error", err,
)
time.Sleep(nextRetry)
} else {
break
}
}
// Found a node with some Raft peers, stop bootstrap since there's
// evidence of an existing cluster. We should get folded in by the
2016-09-01 06:54:53 +00:00
// existing servers if that's the case, so it's cleaner to sit as a
// candidate with no peers so we don't cause spurious elections.
// It's OK this is racy, because even with an initial bootstrap
// as long as one peer runs bootstrap things will work, and if we
// have multiple peers bootstrap in the same way, that's OK. We
// just don't want a server added much later to do a live bootstrap
// and interfere with the cluster. This isn't required for Raft's
// correctness because no server in the existing cluster will vote
// for this server, but it makes things much more stable.
if len(peers) > 0 {
s.logger.Info("Existing Raft peers reported by server, disabling bootstrap mode", "server", server.Name)
s.config.BootstrapExpect = 0
return
}
}
// Attempt a live bootstrap!
var configuration raft.Configuration
var addrs []string
2017-12-12 00:38:52 +00:00
minRaftVersion, err := s.autopilot.MinRaftProtocol()
2017-02-22 20:53:32 +00:00
if err != nil {
s.logger.Error("Failed to read server raft versions", "error", err)
2017-02-22 20:53:32 +00:00
}
for _, server := range servers {
addr := server.Addr.String()
addrs = append(addrs, addr)
2017-02-22 20:53:32 +00:00
var id raft.ServerID
if minRaftVersion >= 3 {
2017-02-22 20:53:32 +00:00
id = raft.ServerID(server.ID)
} else {
id = raft.ServerID(addr)
}
2018-09-20 00:41:36 +00:00
suffrage := raft.Voter
if server.NonVoter {
suffrage = raft.Nonvoter
}
peer := raft.Server{
2018-09-20 00:41:36 +00:00
ID: id,
Address: raft.ServerAddress(addr),
Suffrage: suffrage,
}
configuration.Servers = append(configuration.Servers, peer)
}
s.logger.Info("Found expected number of peers, attempting bootstrap",
"peers", strings.Join(addrs, ","),
)
future := s.raft.BootstrapCluster(configuration)
if err := future.Error(); err != nil {
s.logger.Error("Failed to bootstrap cluster", "error", err)
2014-06-18 23:15:28 +00:00
}
// Bootstrapping complete, or failed for some reason, don't enter this
// again.
s.config.BootstrapExpect = 0
2014-06-18 23:15:28 +00:00
}
// lanNodeFailed is used to handle fail events on the LAN pool.
func (s *Server) lanNodeFailed(me serf.MemberEvent) {
for _, m := range me.Members {
2017-08-30 17:31:36 +00:00
ok, serverMeta := metadata.IsConsulServer(m)
if !ok || serverMeta.Segment != "" {
continue
}
s.logger.Info("Removing LAN server", "server", serverMeta.String())
// Update id to address map
2017-08-30 17:31:36 +00:00
s.serverLookup.RemoveServer(serverMeta)
}
}